Our K props model prices each plate appearance individually, matching pitcher pitch mix against each batter's specific swing tendencies, then sums to a full K distribution. Books price narrative. We price the matchup.
K props are priced by books around narratives and recent starts. We price them from the matchup, one plate appearance at a time.
Before lineups confirm, we project the expected 9-man batting order using recency-weighted slot frequencies from the last 50 games, with a 7-day half-life. Platoon splits match the starter's handedness. Early-line edge captured before public confirmation.
Each plate appearance gets its own XGBoost prediction. The batter's K-rate and whiff features are constructed by weighting their pitch-type-specific stats by the pitcher's actual arsenal usage. A batter who struggles against sliders gets that weighted appropriately when facing a slider-heavy pitcher. TTO (times through order) is the top predictive feature.
With P(K_i) for every PA in hand, we run Monte Carlo simulations of the full game. Each simulation draws a Bernoulli outcome per PA and sums to a K total. Across thousands of runs, the resulting distribution gives precise P(over/under) at any line.
From lineup projection to a K distribution. Five stages, no black boxes.
K prop pricing starts before lineups are confirmed. We project the expected 9-man batting order using recency-weighted slot frequencies from the last 50 team games, with a 7-day half-life. Lineup changes snap in fast.
Platoon filtering selects only games where the opposing pitcher threw from the same hand as today's starter. Players absent 21+ days are excluded as likely IL or traded. The result is a projected order closely matching what gets confirmed, available the moment early lines drop.
How deep a starter goes changes the K total more than most individual matchup factors. We estimate expected PA count with a Bayesian three-factor model: pitcher's shrunk recent mean (730-day window), team hook factor (manager leash tendency), and opponent OBP/K-rate adjustment.
PA count uncertainty feeds directly into the simulation. A starter with unpredictable depth produces a wider K distribution, preventing the model from being overconfident on pitchers who might get pulled in the 4th or go 8 innings.
Each projected PA is scored independently by XGBoost with its own feature vector. The batter's K-rate and whiff features are built by weighting their pitch-type-specific stats by the pitcher's actual arsenal usage. If the pitcher throws 35% sliders, the batter's slider K rate contributes 35% to that feature. Each PA also carries its own TTO value.
The top SHAP feature is TTO (times through order): K rates drop measurably on the 2nd and 3rd trip through the lineup. A starter projected to face the order 2.7 times has a structurally different K profile than one going 2 times flat.
With P(K_i) scored for every projected PA, we run Monte Carlo simulations of the full game. Each simulation draws a Bernoulli outcome for each PA independently and sums to a K total. Thousands of runs build the full K distribution.
The resulting distribution gives precise P(over/under) at any line without any distributional assumption. The simulation naturally captures variance from PA count uncertainty and the spread of individual batter K probabilities across the lineup.
Books set K prop lines around usage trends and public narratives. An ace who went 9 innings last start gets a high line whether or not the matchup supports it. We price from the NegBin distribution and find edge where narrative and reality diverge.
We require a minimum edge threshold before releasing a K prop. Not every starter has a play. When the book line accurately prices the matchup, we pass. Volume is not the goal, precision is.
Full 2025 season. 877 plays logged, consistent edge from April through October.
The 2026 MLB season is live. A full schedule means 1–3 picks almost every day through October. The edge compounds over volume. Subscribe now and be in from the start.
Get K PropsDaily K prop selections with full pitcher profiles, batter matchup context, and PA-level model output. Released before the market moves.
Plays posted when K prop lines become available, opening day through the postseason. Only plays that clear our minimum edge threshold are released, typically 1-3 per slate depending on the card.
Every release includes the pitcher's pitch-mix-weighted whiff and K features, the opposing lineup's K profiles weighted by arsenal, and the XGBoost per-PA output. Understand exactly why each play was released.
Each card shows the full K projection, the model's P(over) at the book line, and the edge percentage after devigging. No black boxes. The Monte Carlo distribution drives every number we release.
All K prop picks land directly in the Whizard subscriber Discord. Each post includes the full model output card, edge percentage, and context summary.
Daily K prop selections from now through October. Pitcher profiles, batter matchup context, and full K distribution included in every release.